Phase 4 of #273 — the actual switch. Both axum routers now serve
their static surface via `tower_http::services::ServeDir` mounted
as a fallback service, reading the dist path from `HIVE_STATIC_DIR`
(set by Phase 3's NixOS module wiring).
Deletes:
- `hive-c0re/assets/{index.html, app.js, dashboard.css}`
- `hive-ag3nt/assets/{index.html, app.js, agent.css, stats.html,
stats.js, screen.html}`
- The whole `hive-fr0nt/` crate (workspace member dropped, both
hive-c0re and hive-ag3nt drop their `hive-fr0nt.workspace = true`
dep). Its contents now live as `@hive/shared` under
`frontend/packages/shared/`.
Rust changes:
- `hive-c0re/src/dashboard.rs`: remove `serve_index`, `serve_css`,
`serve_app_js`, `serve_shared_js`, `serve_marked_js`,
`serve_favicon` (all six `include_str!` handlers); replace their
routes with a single `.fallback_service(ServeDir::new(static_dir))`
on the router. Fail closed (anyhow::bail) if `HIVE_STATIC_DIR` is
unset or not a directory at startup.
- `hive-ag3nt/src/web_ui.rs`: remove `serve_index`, `serve_css`,
`serve_app_js`, `serve_shared_js`, `serve_marked_js`,
`serve_stats`, `serve_stats_js`, `serve_screen`; same
`fallback_service` pattern. `serve_icon` stays (consumes
`/etc/hyperhive/icon.svg` + `branding/hyperhive.svg` fallback,
neither of which lives under the frontend dist).
- `AgentLink` URLs for stats/screen switched from `/stats` / `/screen`
to `/stats.html` / `/screen.html` since ServeDir doesn't auto-
append the extension and the on-disk filename is the natural URL
post-cutover.
- `Cargo.toml` (workspace): drop `hive-fr0nt` member + workspace
dep, add `tower-http = { version = "0.6", features = ["fs"] }`.
- `hive-c0re/Cargo.toml` + `hive-ag3nt/Cargo.toml`: drop the
`hive-fr0nt.workspace = true` dep, add `tower-http.workspace =
true`.
Docs updated:
- `CLAUDE.md`: file map reflects `frontend/` (was `hive-fr0nt/` +
`assets/`) and the ServeDir/HIVE_STATIC_DIR shape.
- `docs/web-ui.md` 'Shape (shared by both)' section: describes the
ServeDir fallback + bundled-by-esbuild surface, no more
`include_str!` references.
- `docs/terminal-rendering.md`: src paths point at
`frontend/packages/{agent,shared}/src/`; marked is the npm dep,
not vendored UMD.
Validation:
- `cargo check --workspace` — clean (5 warnings, all pre-existing
in `rebuild_queue.rs`, none on changed files).
- `cargo clippy --workspace --all-targets` — clean (11 warnings,
same pre-existing source).
- `cd frontend && npm run build` from the prior commit's lockfile
produces the dist directories the new routers consume:
dashboard: `dist/{index.html, static/{app.js, dashboard.css}}`
agent: `dist/{index.html, stats.html, screen.html,
static/{app.js, stats.js, agent.css}}`
(favicon.svg lands in dashboard/ during the nix build —
`nix/frontend.nix` install phase copies `branding/hyperhive.svg`
there, since it's outside the npm tree.)
Refs #273.
The src/index.html / src/stats.html files reference assets at URLs
like /static/app.js, /static/dashboard.css. The initial Phase 1 build
flattened everything to dist/{app.js, dashboard.css, ...} which would
have forced the Phase 4 Rust ServeDir mount to do URL rewriting just
to make the existing HTML references resolve.
Rework: bundles now write to dist/static/, HTML stays at dist/ top
level. Layout matches the URLs the HTML uses, so the Phase 4 mount
is the simplest possible `fallback_service(ServeDir::new(dist))`.
No source-file changes — just the esbuild outfile/outdir paths.
Rebuilt; verified asset filenames + sizes unchanged.
Refs #273.
Phase 3 of #273. Container plumbing for the bundled frontend dist:
- flake.nix overlay: `pkgs.hyperhive-frontend` exposed for the
agent / manager containers (mirrors the existing `pkgs.hyperhive`
pattern); module argument `hyperhiveFrontend = system: self
.packages.${system}.frontend` threads the package into the host
hive-c0re module without forcing operators to apply the overlay
on their host pkgs.
- `services.hive-c0re.frontend` option: pinned to the flake's
frontend package by default, overridable for custom dashboard
SPAs. The hive-c0re systemd service gets `HIVE_STATIC_DIR =
${cfg.frontend}/dashboard` — the Rust binary will pick it up
in Phase 4.
- `hyperhive.frontend.dist` option: per-container, defaults to
`pkgs.hyperhive-frontend`. Override to ship a fully custom
agent SPA (advanced; the default + extraFiles flow handles the
common 'add files' case).
- `hyperhive.frontend.extraFiles` option: attrsOf submodule
(mirroring the `hyperhive.extraMcpServers` shape per damocles'
request so existing #322-style assertions keep their grip).
Each entry has `source` (path relative to agent.nix) and
`target` (URL/disk prefix within the merged static tree,
defaulting to the attribute name). Operator-named example:
the bitburner agent drops `bitburner-dist` into
`/bitburner/` alongside the default agent UI at `/`.
- `hyperhive.frontend.mergedDist` (readOnly): the runCommand
derivation that composes `agent/` from the default dist plus
every `extraFiles` entry. Aborts on overwrite so a filename
collision becomes a build error rather than a silent dist swap.
agent-base.nix + manager.nix set their respective systemd
service `HIVE_STATIC_DIR` to this merged path.
Until Phase 4 lands, the env var is set but unused — the Rust
binaries still serve assets via `include_str!`. The cutover
happens in the next commit on this branch.
Refs #273.
Phase 2 of #273. Adds `packages.${system}.frontend` to the flake —
a `buildNpmPackage` derivation that consumes the lockfile committed
in the previous step and produces two static dist trees under $out:
$out/dashboard/ the hive-c0re dashboard SPA assets
(index.html, app.js, dashboard.css, favicon.svg)
$out/agent/ the per-agent default UI assets
(index.html, app.js, stats.html, stats.js,
agent.css, screen.html)
The dashboard favicon lives outside the frontend src tree
(branding/hyperhive.svg at the repo root). It's passed in as a
callPackage argument so the hermetic build can grab it.
`npmDepsHash` is set to `lib.fakeHash` — the build will fail on
first attempt with the actual sha256 printed; copy that in. Use
`nix run nixpkgs#prefetch-npm-deps -- frontend/package-lock.json`
to recompute locally without a build round-trip (works from
operator's host; iris's container can't recompute it without
prefetch-npm-deps in PATH).
The Rust crates and NixOS modules continue to use the legacy
include_str! routes; cutover happens in Phase 4.
Refs #273.
Follow-up to 9e558c3. Runs `npm install` with the new nodejs_22 + npm
toolchain that just landed in iris's container (approval dfae406),
which generates the lockfile + node_modules tree. Only the lockfile
is checked in; node_modules/ stays in .gitignore.
Pinned versions (resolved by npm from the package.json constraints):
- chart.js 4.4.4 (replaces the jsDelivr CDN script on stats.html)
- marked 4.3.0 (replaces hive-fr0nt/assets/marked.umd.js)
- esbuild 0.25.5 (bumped from 0.24.0 to clear an audit warning
about the dev-server CSRF advisory; bundling
behaviour is unaffected)
Validated locally:
npm install — 0 vulnerabilities reported
npm run build — both workspace builds succeed
dashboard: dist/{app.js (149kb), dashboard.css (33kb), index.html}
agent: dist/{app.js (114kb), stats.js (435kb), agent.css (16kb),
index.html, stats.html, screen.html}
Stripped-comment diff of dist/dashboard.css vs the runtime concat
(BASE_CSS + TERMINAL_CSS + assets/dashboard.css) shows only
whitespace + comment-strip differences — selectors/properties match.
Hermetic-build wiring (the Nix `buildNpmPackage` derivation that
consumes this lockfile) lands in Phase 2 on a follow-up commit.
Refs #273.
Phase 1 of the backend/frontend code split (#273). Additive — no
existing code is touched; the legacy hive-c0re/assets, hive-ag3nt/
assets and hive-fr0nt/assets trees stay in place until the Rust
cutover later in this branch.
Layout:
frontend/package.json npm workspaces root
frontend/packages/shared/ @hive/shared
src/{base,terminal}.css + terminal.js (ES module)
src/index.js re-exports terminal.js
frontend/packages/dashboard/ @hive/dashboard
src/{index.html, app.js, dashboard.css} ported from hive-c0re/assets
build.mjs esbuild config → dist/
frontend/packages/agent/ @hive/agent
src/{index,stats,screen}.html + agent.css
+ {app,stats}.js ported from hive-ag3nt/assets
build.mjs esbuild config → dist/
Changes vs the existing assets:
- terminal.js is an ES module exporting { create, linkify } instead
of assigning to window.HiveTerminal. The dashboard / agent app.js
files re-expose them on window so the IIFE bodies keep working
unchanged through Phase 1; the global aliases can be dropped in a
follow-up once the IIFEs are unwrapped.
- marked is imported from the marked@4.3.0 npm package (replacing
the vendored hive-fr0nt/assets/marked.umd.js bundle).
- chart.js is imported from chart.js@4.4.4 (replacing the jsDelivr
CDN script tag on the per-agent stats page — page now works
offline / on operator machines without internet egress).
- dashboard.css and agent.css both gain @import lines at the top
that pull base.css + terminal.css from @hive/shared, replacing
the runtime string concatenation in serve_css.
- index.html / stats.html collapse from three / two script tags to
one type="module" tag pointing at the bundled output.
package-lock.json is intentionally omitted from this commit — npm
isn't available in the iris container yet (approval pending) and the
lockfile will land in the next commit on this branch once the
toolchain is in place. The PR will not be opened until it's there.
Phase 2 (nix derivations), Phase 3 (container plumbing + the
hyperhive.frontend.extraFiles option for per-agent layering), and
Phase 4 (Rust cutover to tower_http::ServeDir, delete hive-fr0nt
+ legacy assets dirs) land as follow-up commits on this same
branch.
Refs #273.
`.container-icon` had `align-self: stretch` + `aspect-ratio: 1`, so
the square's width tracked the body's height. As soon as state pills
(rate-limited / needs-login / needs-update / ctx) wrapped the head
row, the body grew taller and the icon grew with it — two cards
with different state ended up with visibly different-sized icons
(issue #344).
Pin the icon at 5em; height follows from aspect-ratio. Card-level
`align-items` drops to flex-start so a row taller than the icon
doesn't stretch the icon back out. The card body still flows
however many lines it needs.
Closes#344.
The `⏳ MM:SS` chip on an asked-with-timeout question was rendered
once and then frozen — the operator saw stale info (e.g. 48s
sitting unchanged for the whole TTL window) (issue #335).
Stamp the deadline onto the chip as `data-deadline` and run a
single page-wide setInterval that refreshes every `.q-ttl[data-
deadline]`'s textContent each second. No re-render of the
questions section; no new state on the client. No-op when no
chips are on screen.
Also pulls the bucketed seconds-to-string logic into a
`formatTtl` helper so the renderer and the ticker share one
source of truth.
Closes#335.
The per-container nav strip's <a> elements had class "meta nav-link".
`.container-row .head .meta { margin-left: auto }` then matched every
link, so as flex siblings the first one absorbed all the available
space and the rest packed against it on the right — the icons looked
like they overlapped (issue #333).
Drop `meta` from the link class. Add a `.nav-strip` rule that is
inline-flex with a 0.35em gap so the icons sit on a fixed cadence
regardless of how many backend-supplied links land. Give .nav-link a
real hit target (0.15em / 0.35em padding) + a subtle hover so the
icons read as interactive.
dashboard.rs had the same 12-attempt cap shape as the per-agent
bind_with_retry. Apply the same fix — retry forever with the 2s-capped
backoff, WARN early then INFO once we're clearly stuck on a stale
socket, INFO on success when we did have to retry. Mirrors the
agent change in this PR.
The retry was capped at 12 attempts (~20s of exponential backoff
capped at 2s). Two back-to-back nspawn restarts in #324 left the
previous socket holding the port longer than that budget; once the
cap fired, the web-UI task returned an error and silently died for
the rest of the process lifetime — the agent kept running fine
otherwise (MCP, turn loop), but the operator's dashboard click
hit nothing.
Genuine port collisions are preflighted host-side
(lifecycle::{spawn,rebuild}) and surfaced as a port-conflict banner,
so at this layer a persistent AddrInUse always reflects a
recoverable stale socket. Drop the cap, keep retrying forever with
the same 2s-capped backoff. WARN for the first dozen attempts so a
normal restart-race is visible; INFO after that to avoid spamming
the journal during a long stale-socket hold. Logs a one-line INFO
on success when we did have to retry, so post-mortems can find the
attempt count.
Closes#324.
The previous take put a shared NavLink wire type in hive-sh4re and
duplicated the link-building logic across crates. Per @mara on #326:
that doesn't fit the eventual frontend/backend split goal (#273).
The agent backend is the natural source of truth for what links its
own page exposes; hive-c0re just passes the list through to the
dashboard.
* hive-ag3nt/src/web_ui.rs: agent_links now also serves the
config-repo link + reads agent-declared dashboardLinks extras
from {state_dir}/hyperhive-dashboard-links.json. AgentLink gains a
kind enum (Container | Forge | External) so the frontend can build
the right href no matter which surface is rendering. The host
header is no longer used — URLs are paths for Container/Forge,
absolute for External.
* hive-c0re/src/dashboard.rs: new GET /api/agent/{name}/links route,
a same-origin proxy that fetches the agent's /api/state and
forwards just the links field. No shared wire type — hive-c0re
treats the payload as opaque JSON (serde_json::Value). All failure
modes degrade to an empty list so the dashboard still renders.
* hive-c0re/assets/app.js: container card head row gets an async-
populated icon-only nav strip from the proxy. The hardcoded stats
link, the standalone config-repo trigger, and the extras block are
gone. The deployed:<sha> chip stays — the agent harness can't know
its own deployed sha, so this chip is how the operator sees what's
live alongside the agent's (root-only) config link.
* hive-ag3nt/assets/app.js: agent page meta-links rendered via
el() / textContent (DOM build) so agent-declared icon / label / url
strings never reach innerHTML. kind-based href resolution mirrors
the dashboard side.
* docs/web-ui.md: dashboard + per-agent sections updated for the new
architecture.
Closes#262.
Per @mara on #328: the hand-rolled encoder was over-cautious. Swap
for base64 = 0.22 from crates.io — a standard, widely-trusted dep,
no maintenance surface to carry. Drops the 15-line encoder and its
two RFC 4648 unit tests.
The 'core' Forgejo user (hive-c0re's identity for commits in
core/meta + agent-configs/*) was showing the default hash identicon.
Adds a one-shot ensure_core_avatar in the ensure_all bootstrap that
POSTs the branding PNG to the admin avatar API and writes a marker
file (CORE_AVATAR_MARKER) so subsequent startups skip the call
(delete the marker to re-upload). Best-effort: a non-2xx is logged
and swallowed, doesn't gate startup.
PNG bytes baked in via include_bytes! from branding/hyperhive.png.
Base64 is hand-rolled (one small image in one cold path, not worth
a new workspace dep) with RFC 4648 §10 test vectors.
Closes#320.
PR #321 changed the model priority to: HIVE_DEFAULT_MODEL (from nix config)
> persisted runtime choice > compiled-in DEFAULT_MODEL. The README now
clarifies that the nix config takes precedence and runtime overrides are
reset on rebuild.
Remove the depth-2 cap in walk_meta_inputs so every fetched input
at every depth is surfaced, not just two levels (issue #275). The
uncapped walk needs a guard: a visited-node set makes it a spanning
tree — each fetched node walked once, at its shallowest path — so
shared subtrees don't re-walk and a cycle can't recurse forever.
A two-pass walk (claim a node's direct inputs before descending)
keeps shallow inputs at a shallow path.
Frontend: renderMetaInputs indents each row by its slash-path depth
and shows the leaf segment (full path on hover), plus a select-all /
select-none control so a long input list isn't ticked box by box.
the stream-json result event carries modelUsage.<model>.contextWindow
which is the actual per-inference active window the model enforces.
for claude-sonnet-4-6 this is 200k even though the full prompt cache
can hold millions of tokens via accumulated cache reads.
with the nix-configured sonnet = 1000000 the proactive compact watermark
sat at 750k and was never reached. agents grew context until prompt_too_long
at ~170k — reactive compact, no checkpoint turn.
changes:
- bus gains api_context_window field seeded from modelUsage.*.contextWindow
in each turn's result event. authoritative; falls back to env var, then 200k.
- new effective_context_window(bus) helper used by both watermark functions
- compact_watermark (75%) and auto_reset_watermark (50%) call effective_context_window
- context_tokens() docstring clarified: all three token fields (input +
cache_read + cache_creation) count against the per-inference contextWindow
limit. the large cache_read values seen in the result event are cumulative
across all inferences in a turn, not per-inference.
- /api/state context_window_tokens now reflects the calibrated window
closes#129
extract per-agent forge logic from ensure_all() into sync_agent()
so both the startup sweep and rebuild_agent call identical code.
rebuild now runs: ensure_user_for + ensure_config_repo + push_config
+ meta_read_access + ensure_meta_remote — same as the boot sweep.
missing tokens and drift in any forge state are fixed by rebuild,
not just hive reboot.
if forge_after_first_spawn fails transiently on first spawn the
token is missing. rebuild_agent now calls ensure_user_for so
a manual rebuild (or the startup auto-update scan) recovers
the missing token — no full hive reboot needed.
post_meta_update returns 200 immediately and runs the nix flake
update + agent-rebuild ripple in a background task, so the META
INPUTS panel looked idle for the whole multi-minute window (#259).
Track in-flight runs with a Coordinator atomic counter, exposed via
an RAII MetaUpdateGuard held across run_meta_update. Surface it as
the meta_update_running snapshot field plus a MetaUpdateRunning SSE
event (flipped only when the count crosses 0, so concurrent runs
flip the flag once). The panel shows a pulsing in-progress banner
and disables the update button while a run is active.
